Projecting Health Care Cost Trends: Observations on the SOA’s New Model

Financial Accounting Statement (FAS) 106 requires companies to report and accrue their obligations for postemployment benefits — including retiree health plans — for current and future retirees.¹ The rate of growth in the cost of health care benefits, which must be projected well into the future, can be the most significant assumption in calculating the obligation, a present-value item. So the model and assumptions used in these projections are critical.

Currently, most corporate plan sponsors assume that health care costs will grow at 9 percent in the near term. They also assume that the rate of growth will tamp down to 5 percent over the next six years and remain there for the long term. Our focus in this article is on the long-term projections.

Professor Thomas E. Getzen, a health economist at Temple University, developed the Long Term Healthcare Cost Trends Resource Model in consultation with the Society of Actuaries (SOA).² The project was undertaken to give auditors and others a comprehensive and accepted framework for evaluating long-term trend assumptions and gauging their reasonableness relative to other economic variables.

Getzen’s model provides a logical approach to projecting health care cost increases over a long horizon. He also suggests certain assumptions for input variables, which, as discussed further below, are less sensible. This article compares the Getzen model and assumptions with those used in official federal government projections. It also evaluates the model and assumptions using a range of relevant economic and market statistics, and economic logic.

Historically, the growth in per capita health care costs has generally outpaced price inflation and the growth of wages and other economic variables, often by wide margins. Because the accounting standard presumes that retiree health plans will pay benefits to current and future retirees far in the future (up to 80 years for a young worker), the projected rate of growth in health costs needs to make long-term sense. That said, most private employers have imposed caps on their contributions or make retirees pay the entire premium, making the growth assumption less critical.³ However, even for capped plans, projected medical inflation is important, for example, in modeling levels of future participation and “dropout” rates. While the model should be empirically informed, it should not generate implausible scenarios, such as health care spending’s eventually consuming 100 percent of national income.

The Society of Actuaries/Getzen Model
Starting with an assumption for the percentage of national income spent on health care in the initial year of the long-range projection (presumably somewhat larger than the percentage for the most recent year), the model projects that percentage in future years by applying the following fully adjustable assumptions:

1. Price inflation (in the general economy)
2. Real income growth (inflation-adjusted per-capita national income)
3. Income multiplier (income effect on demand for health care services)
4. Additional medical trend (the effects of technology and other cost-raising factors expected to continue in the future)

The model converts the projected percentage of national income spent on health care to a projection of the annual rate of growth of health care costs.

The user can allow the percentage of national income spent on health care to grow according to the unconstrained model and assumptions, or subject it to one or both of the following constraints:

1. “Resistance point” (percentage above which the economy resists continued growth in health costs, presumably to make room for other economic needs and desires)
2. “Limit year” (point in time at which further health cost increases are largely limited to growth in per capita national income)

The model implements these two constraints smoothly rather than as sharp cutoffs. For example, the resistance point may be set at 25 percent, but then health care continues to grow and take up more of national income, albeit at a slower pace than before the resistance point was met. Similarly, growth does not come to an abrupt stop in the limit year — rather it tapers off over up to 10 years.

Getzen suggests values for each assumption and for the resistance point and limit year. Based on these values, the growth of health care costs increases by 5.2 percent in 2080, after declining gradually from 7.2 percent in 2012, the first year of the long-range projection. This trend curve is higher and longer than that typically used by large retiree health plan sponsors today.

Observations About the Model
The basic Getzen model is logical and represents one fairly reasonable approach to the long-range projection problem. It is roughly similar to the framework currently used in Medicare’s annual trustees’ reports for years beyond the first few of the projection period. The trustees’ framework, in turn, is based largely on the conclusions reached by a 2000 technical review panel of economists and actuaries, namely that projections of future health spending growth should be based on past experience, but should also disregard the impact of one-time historical events such as large expansions in insurance coverage since World War II.

The trustees’ framework has been summarized as “GDP + 1” (where 1 equals technology-induced cost increases). This projects the future annual growth rate of real per capita health care spending as equal to the annual growth rate of per capita gross domestic product plus 1 percentage point for the first 75 years of the projection period. This is currently equivalent to assuming an income multiplier for health spending of 1.0 (and the current trustees’ per capita GDP growth assumption of 1.7 percent) and an extra trend due to technology and other factors of 1.0 percent.⁴

More recently, the trustees have overlain on GDP + 1 a more complex model, which essentially increases the growth rate slightly in the earlier years and decreases it in the far later years. The aim is to capture the prospect of eventual cost containment measures arising from the private sector that will meet with some overall success. For years beyond the traditional 75-year horizon, real per capita health spending is assumed to grow at “GDP.” To arrive at nominal (i.e., non-inflation-adjusted) growth, a general inflation rate (the change in the GDP deflator⁵) must also be assumed and included in the model.

The Getzen model is unique in its use of a resistance point when the health share in GDP exceeds a set, somewhat arbitrary, ratio, to start a formulaic slowing in the projected rate of health spending growth. As mentioned above, the trustees employ a more complex overlay model to come up with a glide path. Others, such as the Treasury Department, have proposed basing the “slowdown” on the common-sense notion that health spending will be contained when it begins to crowd out nonhealth spending (either in the private sector or overall), that is, when broad and long-standing expectations of a steady improvement in standards of living are not met. Presumably stern steps will then be taken to control health spending.⁶ For example, the expansion of HMOs and strict Medicare cost controls in the mid-1990s occurred when excessive health spending growth in the late 1980s began to significantly impinge upon spending in other segments of the economy.

Observations About the Suggested Assumptions
The assumptions proposed by Getzen are somewhat removed from the mainstream of the professional and official consensus and, moreover, are somewhat inconsistent with economic and market statistics and studies (although within a broad range of plausibility).

Getzen assumes a baseline inflation of 3.2 percent for change in the GDP deflator. This is considerably higher than the trustees’ assumptions of 2.8 percent for change in the CPI and 2.45 percent for change in the GDP deflator, which are themselves based on an analysis of historical data (see Figure 1 for a plotting of the inflation data for the last 20 years). Most professional economic forecasters also peg the long-range (CPI) inflation forecast in the range of 2.6 percent.⁷ This, in turn, is consistent with the current implicit CPI inflation market forecast arrived at by comparing yields on nominal and inflation-indexed Treasury securities of the same maturity (see Figure 2 for a plotting of the data of the yield differences).

Because the CPI generally rises a bit faster, a sensible projection for the GDP deflator is closer to 2.3 percent, assuming a CPI-GDP deflator spread of 0.3 percent. Concern about the Federal Reserve’s willingness to control inflation could raise this projection, perhaps by as much as 50 basis points, but a vigorous Federal Reserve response to inflation could also lower the projection.

The market forecast of inflation is particularly relevant as a “feed” component in calculating retiree health liability because the discount rate used for the liability must be based on current conditions in the nominal corporate bond market, which, by arbitrage, must in turn reflect the market’s forecast of inflation.

Getzen assumes per capita income growth of 1.9 percent. As noted above, the trustees’ assumption is 1.7 percent. The trustees arrive at this key assumption for both Social Security and Medicare reports using a building block approach of several components, where component historical data are explored and the relevant underlying trends are projected going forward. This lower assumption has been approved by several technical panels and by the trustees’ staff working groups and, moreover, is consistent with the long-range assumption used by the Congressional Budget Office. Pessimism about future productivity growth would lower this assumption. Broad technological breakthroughs, such as in the production or use of energy, would raise it.

The Getzen assumption for the income multiplier is 1.4. There is a wide range of study estimates of income multipliers, from 0.2 to 0.4, based on household-level data, and 1.0 and above, based on cross-country comparisons. Both methodologies have conceptual strengths and weaknesses, so the 2000 Medicare technical review panel, mentioned above, “split the difference” and used 1.0 as the most reasonable assumption for the income multiplier.

The Getzen assumption for extra trend due to technology and other unknown residual factors is 1.2 percent. After subtracting out the income effect described above and one-time historical factors, the Treasury Department arrives at different answers, ranging from 0.4 (for 1983-2002) to 0.8 (for 1966-2002) to 1.2 (for 1940-2002). In any case, Getzen’s assumption of 1.2 is toward the high end of the range, especially if more recent experience is considered more relevant for projections of the future.⁸

Using the Getzen model with assumptions derived from an eclectic but sensible mix of factors from the bond market, Treasury Department analysis and trustees’ reports, the initial rate of growth of health care costs is 4.9 percent, that is, showing slowing health care growth. Recall that using the model with Getzen’s assumptions arrived at 7.2 percent for the initial rate of growth.

In Table 1, panels A and B show various model assumptions and the resulting rates of growth in health care projected by the model, respectively. The first column in the upper panel has Getzen’s assumptions. The second column has the “slowing health care growth” assumptions (alternative I) with lower price inflation, lower excess health care growth, a low resistance point of 20 percent and a lower initial share of 17 percent. In the first column, the rate of growth declines to 5.2 percent by 2080, while in the second column it declines to 4.3 percent.

Alternative II uses assumptions even more closely tied to those used by the Medicare trustees, and its projections are quite similar to those currently used by many large retiree health plans. Alternative III has lower income growth and higher excess health growth and resistance point than Getzen shows, and its results fall somewhere between Getzen’s and the trustees’. Finally, alternative IV shows high inflation, income growth and taste for health care. Unsurprisingly, its health cost growth rate starts high and remains fairly high throughout the projection horizon.

Table 1
Rates of Health Care Cost Growth
Assumptions for and Results From the SOA/Getzen Model

Panel A: Key Assumptions	Getzen Assumptions	Alternative Assumptions
		I	II	III	IV
2011 Income share	17.5%	17.0%	17.5%	17.5%	17.5%
Inflation	3.2%	2.3%	2.5%	3.0%	3.2%
Income growth	1.9%	1.7%	1.7%	1.5%	2.1%
Income multiplier	1.4	1.0	1.0	1.2	1.4
Excess growth	1.2%	0.8%	1.0%	1.5%	1.5%
Resistance share	25.0%	20.0%	30.0%	40.0%	40.0%
No growth year	2075	2075	2080	2060	2070
Panel B: Results
2012	7.2%	4.9%	5.3%	6.4%	7.8%
2015	7.1%	4.9%	5.3%	6.3%	7.7%
2020	6.9%	5.0%	5.2%	6.2%	7.4%
2025	6.8%	5.0%	5.2%	6.1%	7.3%
2030	6.7%	4.8%	5.1%	6.0%	7.1%
2040	6.2%	4.8%	5.1%	5.8%	6.9%
2050	5.9%	4.7%	5.0%	5.7%	6.7%
2060	5.8%	4.6%	4.9%	5.6%	6.5%
2080	5.2%	4.3%	4.9%	4.6%	5.4%

Source: Watson Wyatt Worldwide.

Conclusions
The Getzen model is a reasonable method to use in calculating FAS 106 obligations. Care must be taken, however, in selecting assumptions for use as variables in the model. Alternatives to those suggested by Getzen — such as assumption sets I and II in this article — may better reflect the empirical evidence and the views of experts, especially in the federal government agencies responsible for doing long-range projections.

¹ Government Accounting Standards Board Statements 43 and 45 impose a similar requirement on state and local governments. The accounting requirements imposed on companies for pension benefits are covered separately by similar standards.
² “Modeling Long Term Healthcare Cost Trends,” Society of Actuaries, December 2007.
³ According to Watson Wyatt’s COMPARISON™ database, in 2006/2007, 40 percent of plans provided by corporations to salaried workers had retirees pay the full cost, 21 percent had caps and 39 percent placed no limits on future increases in employer contributions. Capped benefits are much less common for union plans and for state and local government plans.
⁴ The income multiplier measures the effect of changes in consumer income on the quantity of a good bought. For example, if the quantity of a good purchased increases by 20 percent in response to a 10 percent increase in consumers’ income, the income multiplier is 2.0.
⁵ The increase in the U.S. Consumer Price Index (CPI) is admittedly more familiar to most laypersons. Most economists, including those at the Federal Reserve System, however, prefer to employ the price deflators from the National Income and Product Accounts used in measuring real — that is, inflation-adjusted — quantities of goods and services produced or consumed. Price deflators are considered more accurate than the CPI, because the CPI is based on a fixed basket of goods and services that is adjusted only periodically, whereas the deflators respond quickly to changes in producers’ and consumers’ buying habits as relative prices and incomes change.
⁶ See Jason D. Brown and Ralph M. Monaco, “Possible Alternatives to the Medicare Trustees’ Long-Term Projections of Health Spending,” Department of the Treasury, Office of Economic Policy, Technical Working Paper, 2005, at http://www.treas.gov/offices/economic-policy/papers/healthspending.pdf.
⁷ Most recent conclusion of the Survey of Professional Forecasters, compiled by the Federal Reserve Bank of Philadelphia.
⁸ Professor Getzen, in his paper prepared for the Society of Actuaries, states that the income effects plus the excess technology effects should be consistent with actual rates of “excess” health care spending. He also states that these excess rates are measured with reference to income growth, and that, for the 1980-2005 period, this excess comes to 2.1 percent. This is roughly consistent with what the 2000 Medicare technical panel and Treasury staff have found. But the technical panel and Treasury staff measure excess health spending after subtracting out factors, like insurance penetration, administrative expense growth, defensive medicine and so on, that are thought to be unique to the historical period and not relevant for the future.